Date: Thursday, August 20th, 2026

Time: 2PM ET

Location: Montgomery Knight 325 or Teams

Zhiyuan Zhang

Robotics Ph.D. Candidate

Daniel Guggenheim School of Aerospace Engineering

Georgia Institute of Technology

Committee:

Dr. Panagiotis Tsiotras (advisor) – School of Aerospace Engineering, Georgia Institute of Technology

Dr. Kyriakos Vamvoudakis – School of Aerospace Engineering, School of Electrical and Computer Engineering, Georgia Institute of Technology

Dr. Yongxin Chen– School of Aerospace Engineering, Georgia Institute of Technology

Dr. Glen Chou –  School of Cybersecurity & Privacy, Georgia Institute of Technology

Dr. Sarah Li – School of Aerospace Engineering, Georgia Institute of Technology

Abstract:

Existing dynamic-game solvers can achieve promising practical performance by solving the first-order optimality conditions of a GNE. However, these methods face several challenges: they must operate within real-time control budgets, may converge to non-equilibrium saddle points, and typically return only one of several possible local equilibria.

This dissertation develops computational methods addressing these limitations. It introduces structured Newton and residual-descent solvers that exploit temporal sparsity and active constraints; an inertia-based method for efficiently verifying second-order optimality; and structural modifications that reduce attraction to saddle-type stationary points. It further develops an operator-splitting method that exploits near-potential game structure to accelerate equilibrium computation, as well as a particle-based framework for representing multiple equilibria and coordinating among them through Bayesian belief updates. The proposed methods are evaluated through numerical benchmarks and physical multi-vehicle experiments using the BuzzRacer autonomous racing platform, demonstrating real-time game-theoretic planning in tightly constrained and strategically interactive scenarios.